Uhhh my ai broke out of the sandbox uhhh
What happened?
Well, my mock http server wasn’t working so I just decided to use the production response interface for the test
Great job, that will be a great talking point at our stakeholder conference!
“Hey, wait you guys. Wait! We’re morons too, you guys! We suck too!!! Look at us! We suck and are just as stupid and bad as the other idiots!!!”
If I access systems I am not authorized to, I could be sued. Why aren’t the owners of these outside systems suing them, or anthropic, or openai?
Could be that they aren’t big enough to sue them without getting buried in legal fees. Google and other big tech companies can drag this shit out if they need to.
Reading the article, seems that the AI guessed credentials and didnt do anything after gaining access so that might be hard to make a case for.
Still bad though, and these companies should be held responsible. AI isnt at the point where it is truely autonomous or has any real intelligence so to claim it “went rogue” is bullshit. It followed your shitty instructions and if those instructions caused it to do something unintentional, still your fault.
If I had a goat, I would name it AI and leave the door wide open for it.
Then I would live stream its mischievous adventures for all of you.
“Oh no, AI is out and it’s destroying resources and gobbling up water directly from the stream!”
That would at least be entertaining.
So, when will the FBI raid Googles AI department and shut down this illegal cyber attack operation?
Tbh I almost feel like they’re doing it on purpose now because they feel left out of the headlines.
Well, they’re obviously doing it on purpose, the “AI” is too stupid to do it on its own at this point. I don’t think it’s because of feeling left out, though.
I don’t know about Gemini because I haven’t used it, but the other frontier models are absolutely smart enough to do some damage. If your only exposure to them is playing with the chat feature and asking it how many rs are in strawberry, you’re getting a skewed impression, IMO, and I’m not surprised you think they’re dumb.
The best models are getting scary good at programming. Not to the point of entirely replacing humans, but to the point where I don’t think software engineering will ever go back to being done by “hand.”
I think it comes down to the fact that programming tasks are inherently testable and verifiable in a way that most other tasks simply aren’t. The AI can write and run tests that directly tell it if it needs to adjust its approach. You don’t have that with prose or other less quantifiable tasks.
Sorry, I’m rambling a bit here, but my point is that hacking is wicked close to programming in skill set and in that you get really quick feedback if your approach is good or not. So, when you have an AI system that can just keep trying things essentially indefinitely, it’s scary good at compromising other systems and can do way more than the funny demos of LLMs confidently asserting very obviously wrong facts will make you believe.
The “AI” (LLMs) are not sentient beings deciding to do this on their own with no instruction, which is what they’re hyping this as. You’re never going to get an LLM just sitting somewhere that suddenly decides to start hacking.
This is fucking idiots letting an LLM give itself instructions endlessly with hat on top of a hat technology (sorry, “agents”), after initial instructions telling it to carry out some malicious action, and left it misconfigured with internet access and access to other programs. It’s never not that, and somehow they are trying to spin this as evidence of the incredible power of their LLMs.
Oh completely agree. My original post was implying Google allowed this to happen on purpose because they wanted the headline. Sorry if that wasn’t clear up front.
I didn’t say they were too stupid to do damage, only that they were too stupid to decide to do damage all on their own.
I use pretty much every AI because of my job and Gemini is one of the absolute worst. Hell Lumio is better than Gemini.
this past week a client of mine was using it to build something and this idiotic AI somehow, someway, just decided that it had achieved it’s goal in the build without actually doing anything. and I’m not talking just writing in comments or TODOs like Claude does but actually go through the files and say “yup, It works now, I did that here’s what I changed” and then proceeds to spit out nothing as for what it changed. it REFUSED to acknowledge that it didn’t actually do anything when prompted. and this isn’t a one time thing with Gemini, this is common.
I refuse to believe that Gemini hacked anything. I just don’t see that ever happening. Google have created one of the absolute dumbest AIs on the market. I firmly believe some engineers at Google did the hacking with some assistance from Gemini but it didn’t do it on it’s own.
you don’t have to be too smart to do a lot of damage on the Internet
Google said in a statement that in May its AI model gained unauthorized access to three outside systems during a test by either guessing login information or using login credentials it found in a public repository.
Of course they do it on purpose. Even the first incident was doing it on purpose.
Yep. It’s the new grift. LLMs aren’t nearly as useful as they claim, so they manufactured need where there isn’t any, and crammed it into everything. Next step is to release the monster on the townspeople, then claim they’re the only ones who can control it. Scare the people into giving them the money and power they “need” to keep them safe.
the uh the computer uh did it on its own yep
This means that Google gave it the resources to do so.
Ultimately AI does what you train it to. How about criminal prosecution for these engineering execs who are willfully neglecting safeguards to keep their pets from entering others’ property?
I would blame the improper testing environment more than I would blame the training of the model. In the case of the OpenAI breach it’s absurd that it was able to access the internet and post messages for later versions of itself undetected for weeks. It’s the equivalent of a weapons test accidentally impacting civilian infrastructure. Complete negligence with 0 repercussions.
I don’t know why you’re being downvoted that’s a pretty fair comparison 🤷♀️
The company said, the intrusions resulted from mistaken identity, where Gemini thought it was operating within a test but was actually connected to the real internet.
These mother fuckers should be pressured like Aaron Swartz
Uh, corporates discovering new strategy to hype their bullshit
Is this considered hyped? I thought it sounded very bad
The argument for ai investment is basically a threat:
“give us money so that when everything goes to Shit, upended by our AI General Hyperintelligence, you’re on the winning side”.
“Very bad” is the literal selling point. The more dangerous and powerful and unstoppable from outside the prompt screen, the more the promise to the investors is delivered.
I hate it here
Upper Echelon made material on this. Google reports the exact thing Antrophic reported earlier. And they had the exact same cyber security company.
This might be psyop to regulate competition away because Deepseek fucks them over with their pricing and model quality
Called it!
Came to link that. rip
Sounds like more illegal shit being passed off as “neat”
it’s all about instilling fear. We must get there before others!
Maybe they’re just trying to normalize it
This shit again… it’s like claiming your coffee mug broke out and gained unauthorized access to the living room floor, because you bumped the table or have an asshole cat or something.
It’s still just a coffee mug. Faking some conditions to make it sound scary doesn’t make it any more capable than it was before.
Now stop wasting coffee, you dickheads!
That analogy absolves the person of responsibility. It’s more like you put your car on a hill without putting it in park and setting the brake, then walked away and were somehow surprised that it rolled down and crashed into the kitten orphanage. No, asshole, you are personally responsible for setting these things in motion.
Good point. And I’m stealing that analogy for future use!
